From: Andrea A. <an...@su...> - 2002-03-05 02:02:05
|
On Mon, Mar 04, 2002 at 10:04:51AM -0800, Martin J. Bligh wrote: > High contention on the pagemap_lru lock seems to be a major > scalability problem for rmap at the moment. Based on wli's and > Rik's suggestions, I've made a first cut at a patch to split up the > lock into a per-page lock for each pte_chain. some year ago when I put a spinlock in the page structure I was flamed :) That was fair enough. Sometime the theorical maximum degree of scalability doesn't worth the additional ram usage. but anyways, can you show where do you see this high contenction on the pagemap_lru lock? Maybe that's more a sympthom that the rmap is doing something silly with the lock acquired, can you measure high contention also on my tree on similar workloads? I think we should worry about the pagecache_lock, before the pagemap_lru lock. During heavy paging activity, the system should become I/O bound, and the spinlock there shouldn't matter. while when the system time goes up, it usually doesn't run inside the vm, but it usually runs cached and there the pagecache_lock matters. Andrea |